Evolutionary Intelligence-Based IRS Deployment and Power Allocation for Sum-Rate Enhancement in 6G Networks
TL;DR
Comparative convergence shows that Grid Search provides a near-optimal benchmark at the expense of the highest computational complexity, whereas AO offers a lower-complexity sequential optimization alternative with competitive performance.
Abstract
Sixth generation (6G) wireless networks aim to support high data rates, massive connectivity, and low latency; however, their performance is severely affected by blockage, fading, and interference in practical environments. Intelligent reflecting surfaces (IRS) have emerged as an effective solution for reconfiguring wireless channels and enhancing signal quality. Nevertheless, the benefits of IRS depend strongly on its physical deployment, whereas random placement often leads to inefficient utilization of network resources. This work investigates the optimization of IRS positioning and power coefficients to maximize the achievable sum rate. The proposed framework follows a sequential optimization strategy: first determining the optimal IRS deployment location, followed by power coefficient optimization for IRS-assisted NOMA transmission under quality-of-service constraints. Particle swarm optimization (PSO), real-coded genetic algorithm (GA-RC), binary-coded genetic algorithm (GA-BC), Alternating Optimization (AO), and Grid Search are employed to solve the resulting non-convex optimization problems. These algorithms optimize IRS deployment and NOMA power allocation coefficients, while the IRS phase shifts are assumed fixed and are not adaptively optimized in this work. Simulation results demonstrate that optimized IRS deployment significantly outperforms random placement. At the initial user position, random IRS placement attains a sum rate of 1.10 bits/s/Hz, whereas optimized IRS positioning improves the performance to 3.01 bits/s/Hz using PSO, GA-RC, AO, and Grid Search, and 3.02 bits/s/Hz using GA-BC. Similarly, at a user distance of 500 m, the sum rate with random IRS placement decreases to 0.49 bits/s/Hz, while the optimally positioned IRS maintains significantly higher performance, achieving 3.01 bits/s/Hz with PSO, GA-RC, AO, and Grid Search, and 2.97 bits/s/Hz with GA-BC. Moreover, a 400-element IRS with optimal positioning outperforms an 800-element IRS with random placement, highlighting the importance of deployment strategy over element count. Comparative convergence shows that Grid Search provides a near-optimal benchmark at the expense of the highest computational complexity, whereas AO offers a lower-complexity sequential optimization alternative with competitive performance. Among all methods, PSO achieves the fastest convergence and the lowest computational cost while providing near-optimal performance, making it an effective solution for IRS deployment and NOMA power coefficient optimization in 6G wireless networks.